Exercise :

Over the space $C[a,b]$ we define the norm $$\|f\|_2 = \sqrt{\int_a^bf(x)^2\mathrm{d}x}$$ (i) Show that if the sequence $(f_n)$ converges to $f$ with respect to $\| \cdot \|_\infty$, then it also converges to $f$ with respect to $\| \cdot \|_2$.

(ii) Is $(C[a,b], \| \cdot \|_2)$ a Banach space ?

Attempt :

(i) Let $(f_n)$ be a sequence defined over $C[a,b]$ that converges to $f$ with respect to the norm $\| \cdot \|_\infty$. Then, this means that $\exists n_0 \in \mathbb N$ :

$$\|f_n-f\|_\infty< e \Leftrightarrow \max_{x \in [a,b]}|f_n(x) - f(x)| < \epsilon \; \forall n \geq n_0$$

Now, let's consider the quantity $\| f_n - f \|_2$. This is defined as :

$$\|f_n - f\|_2 = \sqrt{\int_a^b (f_n-f)^2(x)\mathrm{d}x}=\sqrt{\int_a^b[f_n(x)-f(x)]^2\mathrm{d}x}$$

Now, the last expression can be rewritten as :

$$\sqrt{\int_a^b[f_n(x)-f(x)]^2\mathrm{d}x}=\sqrt{\int_a^b |f_n(x) - f(x)|^2\mathrm{d}x}$$

But, it is :

$$|f_n(x)-f(x)|<\max_{x \in [a,b]}|f_n(x)-f(x)| < \epsilon \; \forall n \geq n_0$$

Now, since $|f_n(x)-f(x)| \geq 0$ and $\epsilon >0$, we yield :

$$|f_n(x)-f(x)|<\epsilon \Rightarrow |f_n(x)-f(x)|^2 < \epsilon^2 \equiv \epsilon' \; \forall n\geq n_0$$

And by integrating from $a$ to $b$ since $f_n, f$ are defined over $C[a,b]$ :

$$\int_a^b|f_n(x)-f(x)|^2\mathrm{d}x < \int_a^b\epsilon'\mathrm{d}x=\epsilon'(b-a) \equiv \epsilon'' \forall n\geq n_0$$

Thus, we have $\|f_n-f\|_2 < \epsilon'' \; \forall n\geq n_0$, which means that $(f_n)$ converges to $f$ with respect to the $\|\cdot\|_2$ norm.

Question : Is my approach to (i) correct ? How would one approach (ii) though ?


Part (i) looks fine, for the second part, define the sequence, $$ f_n(x)= \begin{cases} 0,& a\leq x\leq a+\frac{b-a}{2}\\ \frac{2n}{(b-a)}(x-a-\frac{b-a}{2}),& a+\frac{b-a}{2}<x\leq a+ \frac{b-a}{2}+\frac{b-a}{2n}\\ 1,& a+\frac{b-a}{2}+\frac{b-a}{2n}<x\leq b \end{cases} $$ for which it is definitely better to draw a picture (we are taking continuous ramps between $0$ and $1$, with steeper slope, giving the ramp less and less time to get to $1$ as $n$ gets large). The point is that $f_n(x)$ is approximating the discontinuous function $\chi_{[\frac{b-a}{2},1]}(x)$ in $||\cdot||_2$, since $$ ||f_n(x)-\chi_{[\frac{b-a}{2},1]}(x)||\leq \frac{b-a}{4n} $$ where the bound comes from the area of the triangle between the two graphs (everything is $\leq 1$, so squaring only helps us). Since each $f_n$ is continuous and has a discontinuous limit in this metric, $(C([a,b]),||\cdot||_2)$ is not a Banach space.

  • 1
    $\begingroup$ Cool stuff and indeed shows it's not a Banach space BUT how does one come up with all that ? $\endgroup$ – Rebellos Oct 28 '18 at 19:57
  • $\begingroup$ I have tried to add some intuition in the post, but the point is to try and get close in this norm to the simplest discontinuous function with a continuous one. I recommend drawing a picture and then using precalculus to figure out what the $f_n$ in the picture are $\endgroup$ – operatorerror Oct 28 '18 at 19:59

The approach to part (i) is correct. As for part (ii), Consider the sequence of functions $f_{n}(x)=I_{[a,a+\frac{1}{n}]}(x)$ .It converges in $L^2$, but is this sequence uniformly convergent?

  • 1
    $\begingroup$ This isn't an answer to the question. The question is not whether uniform convergence implies onvergence in $L^2$ $\endgroup$ – operatorerror Oct 28 '18 at 19:58
  • $\begingroup$ My mistake, I got carried away thinking about proving the non equivalence of norms. $\endgroup$ – Kesav Krishnan Oct 28 '18 at 20:05
  • $\begingroup$ It happens! Just wanted to point it out $\endgroup$ – operatorerror Oct 28 '18 at 21:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.